Research On Social Search Based On Social Media Mining

In recent years, with the rapid growth in the popularity of Web2.0technologies, Online Social Networks (OSNs) and social network applications, such as Weibo, have undergone explosive growth. This new type of networking platform plays an indispensable part in people’s work and life, while attracting widespread attention from both academic and industry researchers. People are connected to each other through a variety of mutual relationships, forming many large, complicated and content-rich Online Social Networks. Users tend to socialize, communicate, collaborate, share or publish their contents in the OSNs to interact with each other. As a new platform which combines the information sharing and social interaction perfectly, the rapid growth of OSNs has resulted in the substantial amounts of User Generated Content (UGC) generated in Internet. On one hand, it facilitates the flow and dissemination of information. On the other hand, it brings new challenges to the traditional information retrieval method. More and more users tend to seek information through their OSNs rather than the traditional search engine. In this context, social search arises at the historic moment.In this dissertation, we focus on studying and exploring the key issues of social search, based on analyzing the user information and social media data in the OSNs. First, the two basic tasks of social search, which is social search and social recommendation respectively, have been studied. We propose a hybrid social search model, and a collaborative topic prediction model for user interest recommendation. Second, on the basis of analyzing the user attributes and user behaviors in social media, we present an event-based user classification model. Finally, the user influence in social media has been modeled and analyzed. The innovations of this dissertation are mainly reflected in the following aspects:1. Unlike the most search engines, social search is based on user’s online social networks. The crux of social search is to help users find the most accurate information fast and effectively. Users can get not only the information they need, but also the recommendations and the comments on the information sharing from their social friends. In this dissertation, we propose a novel hybrid social search model, which utilizes the user’s social network properties to design the search strategy. Online social networks are organized in accordance with the users, while Web is organized in accordance with the contents. The goal of our paper is to help askers find the appropriate answerers, and give the ranking results of these answerers. During the ranking process on the social network users, the user’s professional degree on a particular field needs to be evaluated. Meanwhile, user’s social relation is another important factor which should be taken into consideration. In addition, user activity and user influence also play important part in calculating the ranking results. Therefore, in order to design the novel hybrid social search model, which enables to provide more accurate information for the users, it is important to consider the topic relevance and user relevance simultaneously, while analyzing and researching on the users’behavior characteristics in online social networks.2. Search and recommendation are two key tasks of social search. Search is an active operation for users to retrieve information, which is based on the analysis of user requirements to provide the personalized search results. On the contrary, recommendation is a passive operation for users, while the system is responsible for recommending the potential interesting users or contents, by analyzing user’s preferences from the historical data. The purpose of this paper is to design a collaborative topic prediction model for user interest recommendation. Topic is the abstract of the User Generated Content (UGC). We build a topic network first to represent the topic relations, and calculate a recommended list of topics according to the correlation between topics. Second, we use the social graph information to recommend the topics that user’s social friends interest in. Finally, the eventual recommendation results are obtained by combining the two recommended methods. In particular, research on user interest recommendation has good application prospects in both academic search and e-commerce fields.3. Weibo media, which combines the information sharing and social interaction perfectly, supports users to respond to social events in a more efficient way. In this paper, we present an event-based user classification model. By analyzing the real world data from Sina Weibo, we investigate the Weibo properties and utilize both content information and Weibo network information to classify the numerous users into four primary groups from the perspective of events:celebrities, organizations/media accounts, grassroots stars and ordinary individuals. Besides, we analyze the behavior characteristics and vocabulary using habits exerted by different categories of users. To classify the users who participate in discussing a particular event, can be beneficial to recommending useful information to the appropriate users. Furthermore, it can be used to help locating the target user groups accurately during the process of social search. The research of event-based user classification also has profound significance for network security, public opinion control and public opinion guide, etc.4. The concept of influence has related definitions in the fields of sociology, communication studies, marketing and political science. In Weibo media, the influence between different users is different. Even for a same user, the influence might be different according to the different events. Therefore, we propose an event-based user influence analysis model in this paper. By analyzing the user attributes, content quality and social relation status in Sina Weibo deeply, we present a definition and computing model for user influence in Weibo media. The experiments results show that our method can identify the top influential users associated with particular events. Analysis of user influence has important significance in expert targeting, authority user mining and opinion leader mining in social search.